smplkit 2.0.2 → 2.0.3

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 18065d9f25f15af5d853a49d46f1263388359c7acaffa765ca7b98e5d2420a98
-  data.tar.gz: 7f386e4671aa2d5e339264007ba46a7b41c9a9cfa47ff1580514aee3248ebd86
+  metadata.gz: 30020aa708d634ffe321a30d974c3f7963eedb3f0c02008d37989bd956a8cefe
+  data.tar.gz: bee093803b8614ab3d36753f59b99ddbc10b012ccd0718f4fdf7b28e1b322baa
 SHA512:
-  metadata.gz: e57e30ebc91651ba5376fc6fde90f5ba4550320470d5e9ae7192b15d4b2e5ce0a5492cdc753f2d04892f897cd7d29422c42c6c3e96008a675c53f1a9dfd24983
-  data.tar.gz: 03eca6e00372adf7cebbdebc719f98ffdc3b2ae0768ba788e1205d29abbeeef8755e32a58a14cee8ac52685274de53ecd15bad4c8c07009fac99f4bd7dcfd132
+  metadata.gz: '0397940b7bfab7871f4f8c9e7e108ba16d82f4f2b7aa444af185a8caa9d095d183e3f7ae011d0e8b45a6d363d3188dcc832cb7268b7ee2f68dbda4a44b548135'
+  data.tar.gz: 356964cbc4b413c5e836866fabb98b335d0c2d8bc4ecbb6f61949a641297bc2b4aa326c52c838e82a075288b57540fd4f56d874dffae923dde261e888a054fbc

data/lib/smplkit/flags/client.rb

@@ -79,6 +79,9 @@ module Smplkit
     # +mgmt.flags.*+. Per-request context is set via
     # +client.set_context([...])+.
     class FlagsClient
+      INITIAL_START_RETRY_DELAY = 1.0
+      MAX_START_RETRY_DELAY = 60.0
+
       def initialize(parent, manage:, metrics:, flags_base_url:, app_base_url:)
         @parent = parent
         @manage = manage
@@ -90,6 +93,9 @@ module Smplkit
 
         @flag_store = {}
         @connected = false
+        @ws_subscribed = false
+        @next_start_attempt_at = 0.0
+        @start_retry_delay = INITIAL_START_RETRY_DELAY
         @cache = ResolutionCache.new
         @handles = {}
         @global_listeners = []
@@ -116,24 +122,44 @@ module Smplkit
 
       # Eagerly initialize the flags subclient.
       #
-      # Drains any pending flag-declaration buffer, fetches all flag
+      # Flushes any pending flag-declaration buffer, fetches all flag
       # definitions, opens the shared WebSocket and subscribes to
       # +flag_changed+ / +flag_deleted+ / +flags_changed+ events.
       #
       # Idempotent — safe to call multiple times. Called automatically on
       # first +flag.get+ evaluation if not invoked manually.
+      #
+      # If the flags-service is unhealthy (e.g. a coordinated rebuild where
+      # the app pod starts before the schema is loaded), the flush or refresh
+      # will fail. Pending declarations stay queued, the client remains
+      # disconnected, and the next call retries after an exponentially
+      # backed-off delay (capped at +MAX_START_RETRY_DELAY+ seconds).
+      # Evaluations during that window fall back to handle defaults.
       def start
         return if @connected
+        return if Process.clock_gettime(Process::CLOCK_MONOTONIC) < @next_start_attempt_at
 
         @environment = @parent._environment
-        flush_flags_safely
-        refresh
+
+        begin
+          @manage.flags.flush
+          refresh
+        rescue StandardError => e
+          schedule_start_retry(e)
+          return
+        end
+
         @connected = true
+        @start_retry_delay = INITIAL_START_RETRY_DELAY
+        @next_start_attempt_at = 0.0
 
         @ws_manager = @parent._ensure_ws
+        return if @ws_subscribed
+
         @ws_manager.on("flag_changed") { |data| handle_flag_changed(data) }
         @ws_manager.on("flag_deleted") { |data| handle_flag_deleted(data) }
         @ws_manager.on("flags_changed") { |data| handle_flags_changed(data) }
+        @ws_subscribed = true
       end
 
       def refresh
@@ -217,10 +243,12 @@ module Smplkit
         handle
       end
 
-      def flush_flags_safely
-        @manage.flags.flush
-      rescue StandardError => e
-        Smplkit.debug("registration", "bulk flag registration failed: #{e.class}: #{e.message}")
+      def schedule_start_retry(exc)
+        delay = @start_retry_delay
+        @next_start_attempt_at = Process.clock_gettime(Process::CLOCK_MONOTONIC) + delay
+        @start_retry_delay = [delay * 2, MAX_START_RETRY_DELAY].min
+        Smplkit.debug("registration",
+                      "flags client start failed (will retry in #{delay}s): #{exc.class}: #{exc.message}")
       end
 
       def fetch_all_flags

data/lib/smplkit/management/buffer.rb

@@ -42,6 +42,10 @@ module Smplkit
     end
 
     # Thread-safe batch buffer for flag declarations.
+    #
+    # Use +peek+ + +commit(ids)+ for the send path so a failed POST leaves
+    # declarations queued for the next attempt. +drain+ is unconditional and
+    # used only by tests/teardown.
     class FlagRegistrationBuffer
       def initialize
         @seen = {}
@@ -61,6 +65,17 @@ module Smplkit
         end
       end
 
+      def peek
+        @lock.synchronize { @pending.dup }
+      end
+
+      def commit(ids)
+        return if ids.nil? || ids.empty?
+
+        committed = ids.to_set
+        @lock.synchronize { @pending.reject! { |item| committed.include?(item["id"]) } }
+      end
+
       def drain
         @lock.synchronize do
           batch = @pending

data/lib/smplkit/management/client.rb

@@ -538,11 +538,24 @@ module Smplkit
 
       def register(declaration)
         @buffer.add(declaration)
-        flush if @buffer.pending_count >= Management::FLAG_BATCH_FLUSH_SIZE
+        return unless @buffer.pending_count >= Management::FLAG_BATCH_FLUSH_SIZE
+
+        begin
+          flush
+        rescue StandardError => e
+          Smplkit.debug("registration", "threshold flag flush failed: #{e.class}: #{e.message}")
+        end
       end
 
+      # POST pending declarations to the bulk endpoint.
+      #
+      # Items remain in the buffer until the request succeeds, so a flush
+      # against an unhealthy service is automatically retried by the next
+      # +flush+ call (lazy +start+ retry, periodic background flush, or
+      # final flush on close). Raises on failure — callers decide whether
+      # to retry.
       def flush
-        batch = @buffer.drain
+        batch = @buffer.peek
         return if batch.empty?
 
         flag_items = batch.map do |entry|
@@ -553,8 +566,11 @@ module Smplkit
         end
         body = SmplkitGeneratedClient::Flags::FlagBulkRequest.new(flags: flag_items)
         ErrorMapping.call { @api.bulk_register_flags(body) }
-      rescue StandardError => e
-        Smplkit.debug("registration", "flag flush failed: #{e.class}: #{e.message}")
+        @buffer.commit(batch.map { |b| b["id"] })
+      end
+
+      def pending_count
+        @buffer.pending_count
       end
 
       def list

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: smplkit
 version: !ruby/object:Gem::Version
-  version: 2.0.2
+  version: 2.0.3
 platform: ruby
 authors:
 - Smpl Solutions LLC